MetaFlux: Global Carbon Flux Estimates from 2001 to 2021

2001 to 2021 global estimates of gross primary production and ecosystem respiration at a 0.25-degree spatial resolution, generated using meta-learning. The dataset was created by Juan Nathaniel of Columbia University to address data sparsity in climate science. Meta-trained models reportedly achieved 9–16% lower validation errors and 4–24% lower errors under extreme conditions compared to non-meta-trained models.

Use Cases

• Modeling global carbon cycle dynamics based on daily and monthly flux estimates.
• Analyzing interannual variability in ecosystem photosynthesis and respiration.
• Benchmarking upscaling algorithms for sparse environmental observations, particularly in tropical and semi-arid regions.
• Studying the correlation between carbon fluxes and solar-induced fluorescence.

Strengths

• Covers a 21-year time range from 2001 to 2021.
• Provides global coverage at a 0.25-degree spatial resolution.
• Reportedly outperforms state-of-the-art upscaling models in critical regions like the tropics and semi-arid areas.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to the underlying in-situ observation network.

Provenance

Source

Columbia University

Collection Method

Generated using an ensemble of meta-trained deep networks (MLP, LSTM, BiLSTM) applied to in-situ observations.

Time Range

2001 to 2021

Geography

Global